Are Websites Going Away? One AI Engineer's Case for llms.txt Replacing Landing Pages

If you are building a company website today, you might be building the last version of it. That is the thesis from someone who spent years on the other side of the screen — building the systems that are replacing how people find information.

John Berryman is the founder of Arcturus Labs and an early engineer on GitHub Copilot. He co-authored the O’Reilly book “Prompt Engineering for LLMs” and has built AI systems ranging from code completion to multi-agent simulations. When he talks about the future of the web, he is not speculating from the sidelines. He is building the tools that accelerate the transition.

The Behavioral Shift Already Happening

The prediction did not come from an abstract thought exercise. It came from a conversation about how people already search. Angelina described watching high school kids on Castro Street pull out their phones and ask ChatGPT what was on sale at Walgreens — not Google, not the Walgreens website.

That story triggered Berryman’s broader prediction. “I don’t think any website is going to exist in five or ten years.”

The reasoning: if consumers are already bypassing websites to talk to AI, and AI can already read structured data and call APIs, then the visual website is becoming a middle layer that nobody needs. The information still matters. The presentation layer does not.

What Replaces the Website

Berryman’s model is specific. Landing pages get replaced by two things: an llms.txt file (a machine-readable markdown document describing everything about your company or product) and an API with tools that AI agents can call.

“lms.txt and an API, which might be an MCP if MCPs are still with it.”

llms.txt, proposed by Jeremy Howard, is essentially a robots.txt for AI agents — a single text file that makes a website’s full content accessible to language models. Instead of scraping HTML and guessing at page structure, agents get clean markdown with the information laid out logically.

The checkout function, the product catalog, the account management — those become standardized API endpoints that AI models are trained to interact with. “The tools for checking out will be standardized. So the models are trained on how to do checkouts.”

The result: your customer never visits your website. Their AI assistant reads your llms.txt, calls your API, and presents the results in a personalized interface.

The PageRank Problem

This raises an obvious question. PageRank works because websites link to other websites, creating a web of authority signals. If websites stop existing, how do you differentiate between competing services?

Berryman acknowledges this is unsolved. “In the future when there is no landing page and nothing to point to” — the current authority system breaks.

His speculation: AI platforms will develop their own scoring systems based on agent interaction quality. When a user’s AI assistant calls your API and the result is good — the user dwells, completes a purchase, expresses satisfaction — that becomes a positive signal. The scoring shifts from static backlinks to dynamic interaction quality.

Hyper-Personalization Kills the 2022 Startup

The implication for startups is severe. If a product satisfies 80% of what a customer wants and costs $100 per month, in a few years that customer will be able to tell an AI to replicate it with their specific preferences — for free or nearly free.

“Instead of paying $100 per month, I have whatever I want immediately. And it’s exactly like what I want.”

Berryman calls this hyper-personalization, and he sees it as the end of a particular startup model. Building a generalized SaaS product and competing on features becomes fragile when the customer can generate a custom version of your product over a weekend of vibe coding.

The startups that survive, in his view, will operate at a higher abstraction level — building infrastructure that AI agents rely on, solving problems that are genuinely novel rather than customizable, and creating value in places where off-the-shelf generation is not sufficient.

FAQ

Will traditional websites still exist in 10 years?

According to John Berryman, landing page websites will be largely replaced by machine-readable data layers — specifically llms.txt files paired with APIs and tool interfaces. The information behind websites persists, but the visual presentation layer becomes unnecessary as AI assistants handle discovery and interaction on behalf of users.

What is llms.txt and why does it matter for businesses?

llms.txt is a proposed standard (created by Jeremy Howard) for making website content machine-readable for AI agents. It is a markdown text file containing a company’s full documentation and product information in a format that language models can parse efficiently. Companies that adopt it early make their content more accessible to AI-powered search and discovery systems.

How will AI change SEO and website discoverability?

Traditional SEO relies on page ranking, backlinks, and keyword optimization for human-readable web pages. As AI agents increasingly mediate between users and information, discoverability shifts toward machine-readable formats (llms.txt), API accessibility, and interaction quality signals — whether the AI agent successfully served the user’s need using your data.

What does hyper-personalization mean for SaaS startups?

When users can instruct AI to replicate 80% of a SaaS product with their specific preferences — essentially vibe-coding a custom version — generalized SaaS products become vulnerable. Startups that compete on features face commoditization. The defensible position shifts to novel problem spaces, infrastructure that AI agents depend on, and value that custom generation cannot replicate.

How does an AI assistant actually search the internet for a user?

The AI uses a search tool (its own index or an existing one like Google’s) to find relevant content, retrieves the most relevant pieces, pastes them into its context window, and synthesizes an answer. The advantage over traditional search: AI has the full conversation context rather than trying to infer intent from five keywords typed into a search box.

What is Arcturus Labs and what AI systems does it build?

Arcturus Labs is an AI consulting firm founded by John Berryman that builds production AI systems across diverse domains. Projects have included multi-agent therapist training platforms for underserved regions, model fine-tuning pipelines that train smaller models to outperform larger ones in specific domains, and RAG architecture consulting for companies moving from demos to production.

Should companies invest in llms.txt files today?

Companies with structured product information or documentation benefit from adopting llms.txt now, even before the full transition Berryman predicts. Making content machine-readable improves visibility to current AI search tools (ChatGPT, Perplexity, Claude) and positions the company ahead of competitors who are still optimizing only for human-readable web pages.

How will product discovery work without websites?

Berryman envisions AI agents reading llms.txt files and calling standardized APIs on behalf of users, presenting results in personalized interfaces. Quality signals will shift from static backlinks to dynamic interaction data — whether users expressed satisfaction, completed actions, or returned. Each user effectively gets their own custom storefront generated by their AI assistant.